File size: 34,095 Bytes
8afb151 ed15028 8afb151 ed15028 8afb151 ed15028 8afb151 ed15028 8afb151 ed15028 8afb151 ed15028 8afb151 ed15028 8afb151 ed15028 8afb151 ed15028 e595317 e2f0d06 ed15028 e595317 ed15028 e595317 ed15028 e595317 ed15028 e595317 ed15028 e595317 ed15028 e595317 ed15028 e595317 ed15028 e595317 ed15028 e595317 ed15028 e2f0d06 ed15028 e2f0d06 ed15028 e2f0d06 e595317 ed15028 e595317 ed15028 e595317 ed15028 e2f0d06 aa15f22 e595317 e2f0d06 ed15028 e2f0d06 ed15028 e2f0d06 ed15028 e595317 ed15028 e595317 ed15028 e595317 ed15028 e595317 ed15028 e595317 ed15028 e595317 ed15028 e595317 ed15028 e595317 ed15028 e595317 ed15028 e595317 ed15028 e595317 ed15028 e595317 ed15028 e595317 ed15028 e595317 ed15028 e595317 ed15028 e595317 ed15028 e595317 ed15028 e595317 ed15028 e595317 ed15028 02b8804 ed15028 48cc8c7 02b8804 48cc8c7 ed15028 02b8804 ed15028 02b8804 ed15028 02b8804 ed15028 48cc8c7 02b8804 ed15028 02b8804 ed15028 48cc8c7 ed15028 48cc8c7 ed15028 48cc8c7 02b8804 ed15028 48cc8c7 ed15028 48cc8c7 ed15028 02b8804 ed15028 48cc8c7 ed15028 02b8804 473ab10 b02956e 473ab10 5a9c33c 473ab10 5a9c33c 02b8804 ed15028 02b8804 ed15028 8afb151 ed15028 8afb151 ed15028 8afb151 ed15028 8afb151 ed15028 8afb151 ed15028 8afb151 ed15028 8afb151 ed15028 8afb151 ed15028 8afb151 ed15028 8afb151 ed15028 8afb151 ed15028 8afb151 ed15028 8afb151 ed15028 8afb151 ed15028 8afb151 ed15028 8afb151 ed15028 8afb151 02b8804 8afb151 02b8804 ed15028 02b8804 ed15028 8afb151 02b8804 ed15028 5a9c33c 5eea840 5a9c33c 707a8d9 b02956e 707a8d9 b02956e 707a8d9 b02956e 5a9c33c ed15028 02b8804 ed15028 02b8804 ed15028 02b8804 ed15028 02b8804 8afb151 ed15028 02b8804 8afb151 02b8804 ed15028 02b8804 ed15028 02b8804 ed15028 02b8804 ed15028 02b8804 ed15028 02b8804 ed15028 8afb151 02b8804 8afb151 02b8804 8afb151 02b8804 8afb151 02b8804 ed15028 8afb151 02b8804 8afb151 02b8804 8afb151 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 | """
Gradio Web UI for Invoice Processing Pipeline
=============================================
Mounted at /web on the main FastAPI app.
"""
from __future__ import annotations
import json
import os
import sys
from typing import Any, Dict, Tuple
import gradio as gr
import httpx
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
_SERVER_URL = "http://localhost:7860"
def _post(path: str, body: Dict[str, Any]) -> Dict[str, Any]:
try:
r = httpx.post(f"{_SERVER_URL}{path}", json=body, timeout=30)
r.raise_for_status()
return r.json()
except Exception as e:
return {"error": str(e)}
def _get(path: str) -> Dict[str, Any]:
try:
r = httpx.get(f"{_SERVER_URL}{path}", timeout=10)
r.raise_for_status()
return r.json()
except Exception as e:
return {"error": str(e)}
# ---------------------------------------------------------------------------
# CSS
# ---------------------------------------------------------------------------
CSS = """
/* ββ global ββ */
body { font-family: 'Inter', sans-serif; }
/* ββ hero banner ββ */
#hero {
background: linear-gradient(135deg, #1e3a5f 0%, #0f2744 50%, #162d4a 100%);
border-radius: 16px;
padding: 32px 40px;
margin-bottom: 8px;
color: white;
}
#hero h1 { font-size: 2rem; font-weight: 700; margin: 0 0 8px 0; color: white; }
#hero p { font-size: 1rem; opacity: 0.85; margin: 0; line-height: 1.6; }
/* ββ agent cards row ββ */
#agent-cards {
display: flex;
gap: 10px;
margin: 12px 0;
flex-wrap: wrap;
}
.agent-card {
flex: 1;
min-width: 130px;
background: white;
border: 1.5px solid #e2e8f0;
border-radius: 12px;
padding: 14px 12px;
text-align: center;
box-shadow: 0 1px 4px rgba(0,0,0,0.06);
}
.agent-card .icon { font-size: 1.6rem; }
.agent-card .name { font-weight: 600; font-size: 0.82rem; color: #1e3a5f; margin-top: 4px; }
.agent-card .desc { font-size: 0.72rem; color: #64748b; margin-top: 2px; }
/* ββ pipeline trace box ββ */
#pipeline-trace textarea {
font-family: 'JetBrains Mono', 'Fira Code', monospace !important;
font-size: 0.82rem !important;
line-height: 1.55 !important;
background: #0f172a !important;
color: #e2e8f0 !important;
border-radius: 10px !important;
}
/* ββ regulator report ββ */
#reg-report textarea {
font-family: 'JetBrains Mono', monospace !important;
font-size: 0.81rem !important;
line-height: 1.5 !important;
background: #0f172a !important;
color: #e2e8f0 !important;
border-radius: 10px !important;
}
/* ββ buttons ββ */
.run-btn { font-size: 1rem !important; font-weight: 600 !important; }
.gr-button-primary { background: #1e3a5f !important; }
/* ββ tab styling ββ */
.tab-nav button { font-weight: 600 !important; font-size: 0.9rem !important; }
/* ββ status pill ββ */
#status-bar textarea {
font-size: 0.82rem !important;
background: #f8fafc !important;
border-radius: 8px !important;
}
"""
HERO_HTML = """
<div id="hero">
<h1>π§Ύ Invoice Processing Pipeline</h1>
<p>
A <strong>self-improving 5-agent system</strong> that generates, extracts, audits, approves,
and regulates invoice fraud β powered by <strong>GRPO-trained Qwen2.5 LoRA agents</strong>.<br>
The Regulator detects Auditor blind spots and biases the Generator toward harder fraud in the next episode,
closing a continuous adversarial improvement loop.
</p>
</div>
"""
AGENT_CARDS_HTML = """
<div id="agent-cards">
<div class="agent-card">
<div class="icon">π―</div>
<div class="name">Regulator</div>
<div class="desc">Detects blind spots, sets weights</div>
</div>
<div class="agent-card">
<div class="icon">β‘</div>
<div class="name">Generator</div>
<div class="desc">Creates adversarial invoices</div>
</div>
<div class="agent-card">
<div class="icon">π</div>
<div class="name">Extractor</div>
<div class="desc">Parses structured fields</div>
</div>
<div class="agent-card">
<div class="icon">π΅οΈ</div>
<div class="name">Auditor</div>
<div class="desc">Flags fraud with confidence</div>
</div>
<div class="agent-card">
<div class="icon">β
</div>
<div class="name">Approver</div>
<div class="desc">Final approve / escalate / reject</div>
</div>
</div>
"""
TASK_DESCRIPTIONS = {
"easy": "Extract structured fields from a single clean invoice.",
"medium": "Clean & normalise a batch of messy invoices (typos, date formats, currencies).",
"hard": "Clean invoices AND reconcile against purchase orders. Flag discrepancies.",
"expert": "Audit invoices for fraud: phantom vendors, price gouging, duplicates, math errors.",
"adversarial": "Extract from an invoice with OCR corruption, fake SUBTOTAL, and FX noise lines.",
"negotiate": "Ask clarification questions, then submit full extraction. Bonus for β€2 questions.",
"supply_chain": "Detect anomalies in delivery records: shortfalls, price spikes, substitutions, phantoms.",
}
PLACEHOLDER_JSON = "// Reset an episode first, then paste or generate JSON here."
# ---------------------------------------------------------------------------
# Pipeline logic
# ---------------------------------------------------------------------------
def _run_pipeline_episode() -> str:
import re as _re
from server.agents import run_extractor, run_auditor
sep = "β" * 58
thin = "β" * 48
lines = [
sep,
" MULTI-AGENT PIPELINE Β· LIVE EPISODE",
sep, "",
]
# ββ STEP 0: Regulator ββββββββββββββββββββββββββββββββββββββββ
report = _get("/regulator/report")
blind_spots = report.get("blind_spots", [])
weights = report.get("generator_weights", {})
lines += [
" π― STEP 0 β REGULATOR",
f" {thin}",
f" Blind spots detected : {', '.join(blind_spots) if blind_spots else 'none'}",
" Fraud weights for next episode:",
]
for ft, w in weights.items():
bar = "β" * max(1, int(w * 24))
flag = " β prioritised" if w >= 0.35 else ""
lines.append(f" {ft:<26} {w:.0%} {bar}{flag}")
lines.append("")
# ββ STEP 1: Generator ββββββββββββββββββββββββββββββββββββββββ
ep = _post("/multi/reset", {})
if "error" in ep:
return f"Error starting episode: {ep['error']}"
episode_id = ep["episode_id"]
n_inv = ep["n_invoices"]
fw = ep.get("fraud_weights_used", {})
raw_text = ep.get("raw_text", "")
ref_data = ep.get("reference_data", "")
dominant = max(fw, key=lambda k: fw.get(k, 0)) if fw else "unknown"
lines += [
" β‘ STEP 1 β GENERATOR (Qwen2.5 LoRA Β· ps2181/generator-lora-qwen2.5-1.5b)",
f" {thin}",
f" Episode : {episode_id[:20]}β¦",
f" Invoices : {n_inv}",
f" Fraud focus : {dominant.replace('_',' ').title()} ({fw.get(dominant,0):.0%} Regulator weight)",
"",
" ββ Invoice Preview βββββββββββββββββββββββββββββββ",
]
for line in raw_text.split("\n")[:14]:
lines.append(f" β {line}")
lines += [" ββ β¦", ""]
# ββ STEP 2: Extractor ββββββββββββββββββββββββββββββββββββββββ
extracted, used_model = run_extractor(raw_text, ref_data)
if not extracted or not used_model:
vendor_m = _re.search(r"Vendor:\s*(.+)", raw_text)
date_m = _re.search(r"Date:\s*(\d{4}-\d{2}-\d{2})", raw_text)
total_m = _re.search(r"(?:Total|TOTAL)[:\s]+[\$Β£β¬]?([\d,]+\.?\d*)", raw_text, _re.IGNORECASE)
vendor = vendor_m.group(1).strip() if vendor_m else "Unknown Vendor"
date = date_m.group(1).strip() if date_m else "2024-01-01"
total = float(total_m.group(1).replace(",", "")) if total_m else 0.0
extracted = {
"vendor": vendor, "date": date, "currency": "USD", "total": total,
"line_items": [{"description": "Office Supplies", "qty": 1,
"unit_price": total, "amount": total}],
}
else:
vendor = extracted.get("vendor", "Unknown Vendor")
date = extracted.get("date", "2024-01-01")
total = extracted.get("total", 0.0)
ext_result = _post("/multi/extract", {"episode_id": episode_id, "extracted_data": extracted})
ext_reward = ext_result.get("reward", 0)
ext_bd = ext_result.get("breakdown", {})
lines += [
" π STEP 2 β EXTRACTOR (Qwen2.5 LoRA Β· ps2181/extractor-lora-qwen2.5-1.5b)",
f" {thin}",
f" Vendor : {vendor}",
f" Date : {date}",
f" Total : ${total:,.2f}",
f" Reward : {ext_reward:.3f} "
f"[format {ext_bd.get('format',0):.2f} "
f"field {ext_bd.get('field_accuracy',0):.2f} "
f"math {ext_bd.get('math_consistency',0):.2f} "
f"completeness {ext_bd.get('completeness',0):.2f}]",
"",
]
# ββ STEP 3: Auditor ββββββββββββββββββββββββββββββββββββββββββ
audit_results, used_model = run_auditor(raw_text, ref_data, n_inv)
inv_ids = list(dict.fromkeys(_re.findall(r"ID:\s*(INV-\d+)", raw_text)))
if not inv_ids:
inv_ids = [f"INV-{i:05d}" for i in range(n_inv)]
if not audit_results:
known = {
"acme corp","globaltech solutions","prime office supplies","datastream inc",
"cloudnine services","metro logistics","pinnacle electronics","summit consulting",
"vertex manufacturing","horizon digital","nexgen software","bluepeak analytics",
}
audit_results = []
for inv_id in inv_ids[:n_inv]:
is_phantom = vendor.lower() not in known
audit_results.append({
"invoice_id": inv_id,
"verdict": "flagged" if is_phantom else "approved",
"fraud_type": "phantom_vendor" if is_phantom else None,
"confidence": 0.78 if is_phantom else 0.85,
})
else:
model_ids = {r.get("invoice_id") for r in audit_results}
if not model_ids.intersection(set(inv_ids)):
for i, r in enumerate(audit_results):
if i < len(inv_ids):
r["invoice_id"] = inv_ids[i]
aud_result = _post("/multi/audit", {"episode_id": episode_id, "audit_results": audit_results})
aud_reward = aud_result.get("mean_reward", 0)
aud_feedback = aud_result.get("feedback", "")
new_report = aud_result.get("tracker_report", {})
lines += [
" π΅οΈ STEP 3 β AUDITOR (Qwen2.5 LoRA Β· ps2181/auditor-lora-qwen2.5-1.5b)",
f" {thin}",
]
for r in audit_results:
v = r.get("verdict", "approved")
ft = r.get("fraud_type")
conf = r.get("confidence", 0)
if v == "flagged":
verdict_str = f"π¨ FLAGGED [{ft.replace('_',' ').upper() if ft else 'UNKNOWN'}] conf={conf:.2f}"
else:
verdict_str = f"β
APPROVED conf={conf:.2f}"
lines.append(f" {r.get('invoice_id','?')} β {verdict_str}")
lines += [
f" Mean reward : {aud_reward:.3f}",
f" Feedback : {aud_feedback[:100]}{'β¦' if len(aud_feedback) > 100 else ''}",
"",
]
# ββ STEP 4: Approver βββββββββββββββββββββββββββββββββββββββββ
approver_decisions = []
for r in audit_results:
v = r.get("verdict", "approved")
conf = r.get("confidence", 0)
if v == "flagged" and conf >= 0.80:
decision = "REJECT"
elif v == "flagged":
decision = "ESCALATE"
else:
decision = "APPROVE"
approver_decisions.append((r.get("invoice_id", "?"), decision))
lines += [
" β
STEP 4 β APPROVER",
f" {thin}",
]
for inv_id, decision in approver_decisions:
icon = {"REJECT": "β", "ESCALATE": "β οΈ ", "APPROVE": "β
"}.get(decision, " ")
lines.append(f" {inv_id} β {icon} {decision}")
n_evaded = sum(1 for r in audit_results if r.get("verdict") == "approved")
gen_reward = 0.85 if n_evaded == n_inv else (0.60 if n_evaded > 0 else 0.10)
lines += [
"",
f" Generator adversarial reward : {gen_reward:.2f}",
f" Invoices that evaded Auditor : {n_evaded} / {n_inv}",
"",
]
# ββ STEP 5: Regulator update βββββββββββββββββββββββββββββββββ
new_blind = new_report.get("blind_spots", [])
new_emerging = new_report.get("emerging_blind_spots", [])
new_weights = new_report.get("generator_weights", {})
lines += [
" π― STEP 5 β REGULATOR UPDATE",
f" {thin}",
f" Episodes recorded : {new_report.get('total_audits_recorded', '?')}",
f" Critical blind spots : {', '.join(new_blind) if new_blind else 'none'}",
f" Emerging blind spots : {', '.join(new_emerging) if new_emerging else 'none'}",
"",
" Generator weights β next episode:",
]
for ft, w in new_weights.items():
bar = "β" * max(1, int(w * 24))
flag = " β BOOSTED" if w >= 0.35 else ""
lines.append(f" {ft:<26} {w:.0%} {bar}{flag}")
lines += [
"",
sep,
" LOOP COMPLETE β weights updated for next episode",
sep,
]
return "\n".join(lines)
# ---------------------------------------------------------------------------
# Regulator report
# ---------------------------------------------------------------------------
def _get_regulator_report() -> str:
data = _get("/regulator/report")
forecast = _get("/regulator/forecast")
calibration = _get("/regulator/calibration")
if "error" in data:
return f"Error: {data['error']}"
sep = "β" * 54
thin = "β" * 44
lines = [
sep,
" REGULATOR DASHBOARD",
sep, "",
f" Episodes recorded : {data.get('total_audits_recorded', 0)} (window = {data.get('window', 30)})",
"",
" DETECTION RATES β improving Β· β declining Β· β stable",
f" {thin}",
]
for ft, status in data.get("detection_rates", {}).items():
lines.append(f" {ft.replace('_',' ').title():<28} {status}")
lines += [
"",
f" False Positive Rate : {data.get('false_positive_rate', 'no data')}",
"",
f" π¨ Critical blind spots : {', '.join(data.get('blind_spots', [])) or 'none'}",
f" β οΈ Emerging blind spots : {', '.join(data.get('emerging_blind_spots', [])) or 'none'}",
"",
" PREDICTIVE FORECAST",
f" {thin}",
]
if "error" not in forecast:
for ft, trend in forecast.get("trends", {}).items():
lines.append(f" {ft.replace('_',' ').title():<28} {trend}")
rec = forecast.get("recommendation", "")
if rec:
lines += ["", f" β {rec}"]
lines += [
"",
" CONFIDENCE CALIBRATION (high conf + wrong = dangerous)",
f" {thin}",
]
if "error" not in calibration:
for ft, cal in calibration.items():
if isinstance(cal, dict) and cal.get("status") != "no data":
lines.append(f" {ft.replace('_',' ').title():<28} {cal.get('status','')}")
wrong = cal.get("mean_confidence_when_wrong")
right = cal.get("mean_confidence_when_correct")
if wrong is not None:
lines.append(f" wrong conf={wrong:.2f} correct conf={right if right else 'N/A'}")
lines += [
"",
" GENERATOR WEIGHTS β next episode bias",
f" {thin}",
]
for ft, w in data.get("generator_weights", {}).items():
bar = "β" * int(w * 30)
flag = " β prioritised" if w >= 0.35 else ""
lines.append(f" {ft.replace('_',' ').title():<28} {w:.1%} {bar}{flag}")
lines += ["", f" Verdict : {data.get('verdict', '')}", sep]
return "\n".join(lines)
_GITHUB_RAW = "https://raw.githubusercontent.com/ps2181/invoice-processing-pipeline/main/assets"
def _make_training_curves() -> str:
"""Return HTML showing the 3 real Colab training curve images from GitHub."""
img_style = "width:100%;border-radius:10px;margin-bottom:16px;"
label_style = (
"color:#94a3b8;font-size:0.82rem;font-family:monospace;"
"text-align:center;display:block;margin-bottom:20px;"
)
return f"""
<div style="padding:8px;">
<p style="color:#94a3b8;font-size:0.85rem;margin-bottom:16px;">
Real training runs on Google Colab A100 β Qwen2.5-1.5B-Instruct + LoRA r=16 + TRL GRPOTrainer + Unsloth
</p>
<img src="{_GITHUB_RAW}/reward_curve.png" style="{img_style}" />
<span style="{label_style}">
π Extractor β Total reward (4 signals) β Β· Live /grader score peaked 0.914 at step 15 Β· Crashed at step 20 due to _MAX_SESSIONS=50 bug (fixed to 200)
</span>
<img src="{_GITHUB_RAW}/auditor_reward_curve_run2.png" style="{img_style}" />
<span style="{label_style}">
π΅οΈ Auditor β Run 2 (bug fixed) Β· Total reward β 0.48β0.72 Β· Live env reward β 0.28β0.52 Β· Run 1 had dead signal (episode_id list bug)
</span>
<img src="{_GITHUB_RAW}/generator_reward_curve.png" style="{img_style}" />
<span style="{label_style}">
β‘ Generator β Fraud plausibility (format) learned ~0.19β0.25 Β· Live evasion reward stuck at 0.01 (same episode_id bug) β needs rerun with fix
</span>
</div>
"""
def _seed_demo_data() -> str:
data = _post("/regulator/demo_seed", {})
if "error" in data:
return f"Error: {data['error']}"
return (
"β
Demo data seeded β phantom_vendor ~32% (blind spot), "
"duplicate_submission declining (emerging)\n\n"
+ _get_regulator_report()
)
# ---------------------------------------------------------------------------
# Agent Tester helpers
# ---------------------------------------------------------------------------
def _call_llm(task_id: str, obs: Dict[str, Any], step: int) -> Tuple[str, str]:
try:
from openai import OpenAI
from inference import SYSTEM_PROMPTS, build_user_prompt, MODEL_NAME, API_BASE_URL, API_KEY
if not API_KEY:
return "{}", "β οΈ No API key found β set HF_TOKEN or API_KEY env var."
client = OpenAI(base_url=API_BASE_URL, api_key=API_KEY)
user_prompt = build_user_prompt(task_id, obs, step)
completion = client.chat.completions.create(
model=MODEL_NAME,
messages=[
{"role": "system", "content": SYSTEM_PROMPTS[task_id]},
{"role": "user", "content": user_prompt},
],
temperature=0.3,
max_tokens=2048,
)
raw = (completion.choices[0].message.content or "").strip()
if raw.startswith("```"):
raw = raw.split("\n", 1)[-1] if "\n" in raw else raw[3:]
if raw.endswith("```"):
raw = raw[:-3]
raw = raw.strip()
parsed = json.loads(raw)
return json.dumps(parsed, indent=2), f"β
LLM ({MODEL_NAME}) responded."
except json.JSONDecodeError as e:
return "{}", f"β LLM returned invalid JSON: {e}"
except Exception as e:
return "{}", f"β LLM error: {e}"
# ---------------------------------------------------------------------------
# Build Gradio app
# ---------------------------------------------------------------------------
def build_ui() -> gr.Blocks:
init_state = {"episode_id": None, "obs": None, "step": 0, "history": []}
def do_reset(task_id: str, state: dict):
data = _post("/reset", {"task_id": task_id})
if "error" in data:
return (
state,
gr.update(value=f"β {data['error']}"),
gr.update(value=""),
gr.update(value=""),
gr.update(value=""),
gr.update(value=PLACEHOLDER_JSON),
gr.update(value=""),
gr.update(value=""),
gr.update(interactive=False),
gr.update(interactive=False),
)
obs = data["observation"]
ep = data["info"]["episode_id"]
new_state = {"episode_id": ep, "obs": obs, "step": 0, "history": []}
status = (
f"β
Episode started task={task_id} id={ep[:12]}β¦\n"
f"Max attempts: {obs['max_attempts']}"
)
return (
new_state,
gr.update(value=status),
gr.update(value=obs["task_description"]),
gr.update(value=obs["raw_text"]),
gr.update(value=obs.get("reference_data") or ""),
gr.update(value=PLACEHOLDER_JSON),
gr.update(value=""),
gr.update(value=""),
gr.update(interactive=True),
gr.update(interactive=True),
)
def do_llm(task_id: str, state: dict):
if not state.get("obs"):
return PLACEHOLDER_JSON, "β οΈ Reset an episode first."
return _call_llm(task_id, state["obs"], state["step"] + 1)
def do_submit(json_str: str, state: dict):
if not state.get("episode_id"):
return state, "β οΈ Reset an episode first.", "", "", ""
try:
extracted = json.loads(json_str)
except json.JSONDecodeError as e:
return state, f"β Invalid JSON: {e}", "", "", ""
data = _post("/step", {"extracted_data": extracted, "episode_id": state["episode_id"]})
if "error" in data:
return state, f"β {data['error']}", "", "", ""
obs = data["observation"]
reward = data.get("reward", 0.0)
done = data.get("done", False)
state["obs"] = obs
state["step"] += 1
state["history"].append(
f"Step {state['step']}: reward={reward:.3f}" + (" β done" if done else "")
)
status = (
f"Step {state['step']} / {obs['max_attempts']} "
f"Reward: {reward:.3f} "
f"{'π Done' if done else 'In progressβ¦'}"
)
bd = obs.get("reward_breakdown")
return (
state, status,
obs.get("feedback") or "No feedback yet.",
"\n".join(state["history"]),
json.dumps(bd, indent=2) if bd else "",
)
def _get_model_status() -> str:
from server.agents import models_status
s = models_status()
if not s:
return "No models checked yet β run a pipeline episode to trigger loading."
return "\n".join(f" {k:<12} {v}" for k, v in s.items())
# ββ Build layout ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
with gr.Blocks(
title="Invoice Processing Pipeline β Multi-Agent Fraud Detection",
theme=gr.themes.Soft(
primary_hue=gr.themes.colors.blue,
secondary_hue=gr.themes.colors.slate,
font=[gr.themes.GoogleFont("Inter"), "sans-serif"],
),
css=CSS,
) as demo:
gr.HTML(HERO_HTML)
gr.HTML(AGENT_CARDS_HTML)
session_state = gr.State(init_state)
with gr.Tabs():
# ================================================================
# TAB 1 β Multi-Agent Pipeline (FIRST β most impressive)
# ================================================================
with gr.Tab("π Live Pipeline"):
gr.Markdown(
"### Watch all 5 agents work together in real time\n"
"Each run is a **fresh episode**: the Regulator reads current blind spots, "
"the Generator creates adversarial invoices biased toward weak fraud types, "
"the Extractor parses them, the Auditor flags fraud, the Approver decides, "
"and the Regulator updates weights for the next round."
)
with gr.Row():
run_btn = gr.Button("βΆ Run Full Pipeline Episode", variant="primary",
size="lg", scale=4, elem_classes=["run-btn"])
model_btn = gr.Button("π Agent Status", variant="secondary", scale=1)
model_status_box = gr.Textbox(
label="Loaded Agents",
interactive=False,
lines=3,
value="Click 'π Agent Status' to see which LoRA models are loaded.",
elem_id="status-bar",
)
pipeline_output = gr.Textbox(
label="Pipeline Trace",
interactive=False,
lines=50,
value=(
"Click βΆ Run Full Pipeline Episode to start.\n\n"
"You will see all 5 agents execute in sequence with live rewards and "
"the Regulator's self-improvement loop closing at the end."
),
elem_id="pipeline-trace",
)
run_btn.click(fn=_run_pipeline_episode, inputs=[], outputs=[pipeline_output])
model_btn.click(fn=_get_model_status, inputs=[], outputs=[model_status_box])
# ================================================================
# TAB 2 β Regulator Dashboard
# ================================================================
with gr.Tab("π Regulator Dashboard"):
gr.Markdown(
"### Cross-Episode Auditor Oversight\n"
"The Regulator monitors detection rates across episodes, "
"identifies blind spots using **predictive trend analysis**, "
"flags **overconfident misses** via confidence calibration, "
"and dynamically reweights the Generator toward under-detected fraud types."
)
with gr.Row():
refresh_btn = gr.Button("π Refresh Report", variant="primary", scale=2)
seed_btn = gr.Button("π± Load Demo Data", variant="secondary", scale=1)
report_box = gr.Textbox(
label="Regulator Report",
interactive=False,
lines=30,
value="Click π Refresh Report or π± Load Demo Data to begin.",
elem_id="reg-report",
)
refresh_btn.click(fn=_get_regulator_report, inputs=[], outputs=[report_box])
seed_btn.click(fn=_seed_demo_data, inputs=[], outputs=[report_box])
# ================================================================
# TAB 3 β Training Results
# ================================================================
with gr.Tab("π Training Results"):
gr.Markdown(
"### GRPO Training Results β Real Data from Colab Runs\n"
"Each agent trained with **TRL GRPOTrainer + Unsloth** on live environment data. "
"The deployed HF Space served as the live reward verifier.\n\n"
"| Agent | Result | Notes |\n"
"|-------|--------|-------|\n"
"| **Extractor** | Live grader score peaked at **0.914** | Crashed mid-run due to session pool bug (_MAX_SESSIONS=50), fixed to 200 |\n"
"| **Auditor** | Live reward climbed **0.01 β 0.52** | First run had dead reward signal (episode_id list bug), fixed and retrained |\n"
"| **Generator** | Fraud format learned (~0.22) | Live evasion reward had same episode_id bug β needs rerun with fix |"
)
try:
_curves_html = _make_training_curves()
except Exception:
_curves_html = "<p style='color:#f87171'>Could not render training curves.</p>"
gr.HTML(value=_curves_html)
gr.Markdown(
"**Model checkpoints:** "
"[Extractor LoRA](https://huggingface.co/ps2181/extractor-lora-qwen2.5-1.5b) Β· "
"[Auditor LoRA](https://huggingface.co/ps2181/auditor-lora-qwen2.5-1.5b) Β· "
"[Generator LoRA](https://huggingface.co/ps2181/generator-lora-qwen2.5-1.5b)"
)
# ================================================================
# TAB 4 β Single-Agent Tester
# ================================================================
with gr.Tab("π§ͺ Single Agent Tester"):
gr.Markdown(
"### Test the OpenEnv environment directly\n"
"Select a task, reset to load an invoice, then submit structured JSON "
"to see the grader score and feedback. Covers 7 task types from easy extraction "
"to supply chain anomaly detection."
)
with gr.Row():
task_dd = gr.Dropdown(
choices=list(TASK_DESCRIPTIONS.keys()),
value="easy",
label="Task type",
scale=1,
)
reset_btn = gr.Button("π Reset Episode", variant="primary", scale=1)
status_box = gr.Textbox(
label="Status",
interactive=False,
scale=3,
lines=2,
elem_id="status-bar",
)
task_info = gr.Textbox(label="Task description", interactive=False, lines=1)
with gr.Row():
with gr.Column(scale=5):
invoice_box = gr.Textbox(
label="π Invoice (raw text)",
interactive=False,
lines=16,
max_lines=30,
)
ref_box = gr.Textbox(
label="π Reference data (PO / vendor registry / catalog)",
interactive=False,
lines=8,
max_lines=16,
)
with gr.Column(scale=5):
json_box = gr.Code(
label="Extracted JSON",
language="json",
lines=16,
value=PLACEHOLDER_JSON,
)
with gr.Row():
llm_btn = gr.Button(
"π€ Run LLM Agent",
variant="secondary",
interactive=False,
)
submit_btn = gr.Button(
"β
Submit",
variant="primary",
interactive=False,
)
llm_status = gr.Textbox(
label="Agent status",
interactive=False,
lines=1,
)
with gr.Row():
feedback_box = gr.Textbox(
label="π Grader Feedback",
interactive=False,
lines=5,
scale=3,
)
breakdown_box = gr.Code(
label="Reward Breakdown",
language="json",
lines=5,
interactive=False,
scale=2,
)
history_box = gr.Textbox(
label="Step history",
interactive=False,
lines=3,
)
task_dd.change(
fn=lambda t: TASK_DESCRIPTIONS.get(t, ""),
inputs=[task_dd],
outputs=[task_info],
)
reset_btn.click(
fn=do_reset,
inputs=[task_dd, session_state],
outputs=[
session_state, status_box, task_info,
invoice_box, ref_box, json_box,
feedback_box, history_box,
llm_btn, submit_btn,
],
)
llm_btn.click(
fn=do_llm,
inputs=[task_dd, session_state],
outputs=[json_box, llm_status],
)
submit_btn.click(
fn=do_submit,
inputs=[json_box, session_state],
outputs=[session_state, status_box, feedback_box, history_box, breakdown_box],
)
return demo
|